🚀 feat(model): add UniNet

[ashwinvaidya17]

📝 Description

• Add UniNet model
• uninet.yaml does not work as Early Stopping from config is broken for all the models.
• Closes 📋 [TASK] Implement UniNet - CVPR 2025 #2781

Results

Combined

category	Original Acc	Original AUROC	Anomalib AUROC	Original F1	Anomalib F1
bottle	100	100	99.76	100	98.39
cable	98.8	97.6	98.18	94.5	93.99
capsule	100	100	94.81	100	95.89
carpet	99.6	98.6	89.61	97.8	89.77
grid	99.2	97.8	100	95	99.12
hazelnut	100	100	99.93	99.3	98.57
leather	100	100	99.39	99.5	97.33
metal_nut	100	100	100	100	99.46
pill	99.7	98.4	96.73	99.3	97.12
screw	100	100	94.14	100	91.83
tile	98.9	96.9	98.48	94.7	95.81
toothbrush	100	100	95.28	100	90.91
transistor	99.2	99.2	98.54	97.5	96.1
wood	99.9	100	99.12	98.4	96.67
zipper	99.7	99	94.01	97.9	95.83

Original

object	image_auroc	acc	f1
carpet	98.6	99.6	97.8
bottle	100	100	100
hazelnut	100	100	99.3
leather	100	100	99.5
cable	97.6	98.8	94.5
capsule	100	100	100
grid	97.8	99.2	95
pill	98.4	99.7	99.3
transistor	99.2	99.2	97.5
metal_nut	100	100	100
screw	100	100	100
toothbrush	100	100	100
zipper	99	99.7	97.9
tile	96.9	98.9	94.7
wood	100	99.9	98.4

Anomalib

	data.init_args.category	image_AUROC	image_F1Score
0	bottle	99.76	98.39
1	cable	98.18	93.99
2	capsule	94.81	95.89
3	carpet	89.61	89.77
4	grid	100	99.12
5	hazelnut	99.93	98.57
6	leather	99.39	97.33
7	metal_nut	100	99.46
8	pill	96.73	97.12
9	screw	94.14	91.83
10	tile	98.48	95.81
11	toothbrush	95.28	90.91
12	transistor	98.54	96.1
13	wood	99.12	96.67
14	zipper	94.01	95.83

Anomaly Map Examples

✨ Changes

Select what type of change your PR is:

• 🚀 New feature (non-breaking change which adds functionality)
• 🐞 Bug fix (non-breaking change which fixes an issue)
• ♻️ Refactor (non-breaking change which refactors the code base)
• ⚡️ Performance improvements
• 🎨 Style changes (code style/formatting)
• 🧪 Tests (adding/modifying tests)
• 📚 Documentation update
• 📦 Build system changes
• 🏗️ CI/CD configuration
• 🔧 Chore (general maintenance)
• 🔒 Security update
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)

✅ Checklist

Before you submit your pull request, please make sure you have completed the following steps:

• 📚 I have made the necessary updates to the documentation (if applicable).
• 🧪 I have written tests that support my changes and prove that my fix is effective or my feature works (if applicable).
• 🏷️ My PR title follows conventional commit format.

For more information about code review checklists, see the Code Review Checklist.

[samet-akcay]

Love the implementation! Only few minor comments

[samet-akcay]

would it be an idea to have a better module name for this? de_resnet.py never made sense to me

[ashwinvaidya17]

Resnet Transpose? Like ConvTranspose. I think de resnet was the name given by the authors of reverse distillation.

[samet-akcay]

• resnet_decoder.py could also work.
• Would it make sense to rename ResNet to ResNetDecoder as well?
• Also the docstring of ResNet should be updated with the correct imports I guess. The imports still show that the module is in reverse distillation

[samet-akcay]

any idea how multi-class and few-show can work here?